Handling forked processes with universalasync decorators

[cheshirex]

I just debugged a case where a user of a module I maintain which uses universalasync had combined it with multiprocessing and caused a hang. I figure the case is atypical, but worth bringing up.

Combining multiprocessing with async doesn't normally make sense, but I'm using universalasync to maintain a library that provides service to both sync and async code, and in this particular case, had sync code running with a ProcessPool, and the called function was in turn calling another function using async_to_sync_wraps. The result was a hang, because the fork still picks up the event loop from the parent process, but that event loop isn't valid in the subprocess.

I've got a local hack fixing this in my own use case now, but the AI-suggested fix within universalasync would be:

_LOOP_PID: int = os.getpid()

def get_event_loop() -> asyncio.AbstractEventLoop:
    global _LOOP_PID
    loop = _get_event_loop()
    current_pid = os.getpid()
    if loop.is_closed() or current_pid != _LOOP_PID:
        _LOOP_PID = current_pid
        return _create_new_event_loop()
    return loop

i.e. check the current process ID against whatever process first entered the code, and if they don't match, we're in a forked process, so get a new event loop and save the process ID in memory for the current process.

Thanks!